|Publication number||US4809209 A|
|Application number||US 06/769,059|
|Publication date||Feb 28, 1989|
|Filing date||Aug 26, 1985|
|Priority date||Aug 26, 1985|
|Publication number||06769059, 769059, US 4809209 A, US 4809209A, US-A-4809209, US4809209 A, US4809209A|
|Inventors||Stanley A. White|
|Original Assignee||Rockwell International Corporation|
|Export Citation||BiBTeX, EndNote, RefMan|
|Patent Citations (12), Referenced by (17), Classifications (5), Legal Events (10)|
|External Links: USPTO, USPTO Assignment, Espacenet|
The present invention relates to discrete-time analog filters, also called sampled-data filters. The invention particularly relates to those filters intended for use with signals having frequencies or sampling rates in the range of several hundred megahertz to a few gigahertz.
A sampled-data filter may be formed of charge-transfer devices, such as charge-coupled devices. One such filter is a bandpass filter. For many applications, the sampling frequency for the signal input to such a filter may be very high, on the order of several hundred megahertz to a few gigahertz. A typical bandpass filter must meet specifications for a number of performance criteria, including passband width, transition width, and stopband rejection. If the filter is an "agile" bandpass filter for which the filter characteristics are changeable, additional performance criteria such as the speed with which changes are effected are also specified. Typically, changes in filter characteristics must be accomplished quickly, within a few cycles of the sampling frequency.
Discrete-time filters have typically been implemented using conventional analog finite-impulse-response (FIR) filters having analog coefficient multipliers, as shown in FIG. 1. A conventional analog FIR filter has an input delay line 10 with a plurality of taps 12, each having a tap weight. Each tap weight is an analog value to be multiplied by the signal tapped from the input delay line. The outputs of the taps 12 are combined in a summer 14 to form the output from the filter. Experience has shown, however, that the tolerance of each analog tap-weight value is difficult to control. Thus, the filter may not be accurately repeatable or reproducible. Further, when the sampling frequency is high, on the order of several hundred megahertz to a few gigahertz, it has been found that it is difficult to multiply the incoming high-frequency signal by these analog-coefficient multipliers and obtain accurate results because of the difficulties of high-speed multiplication. In addition, for an "agile filter", it is difficult to change the analog coefficients of the tap weights quickly, and maintain sufficient accuracy in those coefficients and the multiplication process.
In an attempt to improve somewhat on the performance of the conventional analog finite-impulse-response filter, the use of digital tap weights in association with each tap has been suggested. One accurate method to digitally implement the tap weights is by multiplying digital-to-analog converters; however, this method is restricted to low sampling-rate frequencies.
A structure for digitally implementing the tap weights for higher-speed operation is shown in FIG. 2. In such a finite-impulse-response filter, several binary-analog correlators (BAC's) 20 are connected in parallel. Each BAC 20 is a separate finite-impulse-response filter in which the coefficient multipliers in each tap 22 are the corresponding bits of the several filter coefficients when expressed as binary numbers. Each bit of each filter coefficient is mechanized using one BAC 20. As many BAC's are used as there are bits in each binary filter coefficient. The first BAC filter 20a uses for its tap weights or coefficient multipliers only the most-significant bit of each coefficient. The second filter 20b uses for its tap weights only the second-most-significant bit of each coefficient, and so on. As many of these binary-analog correlators are connected in parallel as there are bits in the multiplier coefficients. The output of each correlator is weighted appropriately in a weighting element 24 to correspond to the significance of the bit of that correlator in the coefficient multipliers. The weighted outputs are then summed in an output summer 26 to obtain the desired filter output.
Each tap multiplier 22 in the BAC filter therefore has a binary-coefficient, resulting in a known accuracy of multiplication for high-frequency signals; but, since the filter requires as many separate binary-analog correlators 20 as there are bits specified in the coefficient multipliers of the filter, these filters generally occupy too much circuit space, consume too much power, are too expensive for most practical systems. In addition, the large number of individual binary-coefficient multipliers, each with its own reliability and tolerance characteristics, may reduce overall system reliability.
In addition, each of the numerous delay elements in each BAC forming the filter has its own charge-transfer inefficiency, further contributing to the performance error for the overall filter. For example, a sampled-data filter may be designed for operation at a sample frequency of 870 MHz, with a 3 MHz passband width, a 95 DB stopband rejection, a transition frequency of 6 MHz, and intended to change frequencies in 6.7 nanoseconds. If each tap weight is approximated as a fourteen-bit binary number, with one GaAs cell representing each bit, that filter may be implemented with a gallium arsenide CCD-filter design with approximately 400 taps, if the charge-transfer-inefficiency variance is kept sufficiently small to keep the accuracy of the filter within acceptable limits. With N=400 taps, B=14 bits/tap, and 1 cell/bit, N times B=5600 cells are required to implement the filter. To keep the overall operation of the filter within any kind of acceptable limits, the variation in the charge-transfer inefficiencies of the cells must be kept to less than approximately 10-6. The yield of devices that meet such performance specifications is small, causing such filters to be prohibitively expensive.
For filtering signals having low sampling frequencies, such as speech, some work has been done on decimated or thinned analog and digital finite-impulse-response filters. These thinned filters have a greatly reduced number of weighting coefficients. An example of a thinned analog filter is shown in FIG. 3. This reduction in the number of filter coefficients is possible because FIR filter tap weights are generally, from an information-theoretic point of view, overspecified. Coefficient values may be removed from the filters, thereby making the corresponding points on the impulse-response curve zero. By interpolation, however, these missing points may be recovered with suitable accuracy. The filter formed by a thinned FIR filter with an interpolator coupled to the output may be viewed as a pair of filters, with the thinned filter connected in tandem with the interpolator. Papers discussing thinned FIR filters are:
1. M. W. Smith and D. C. Farden, "Thinning The Impulse Response of FIR Digital Filters", Proc. ICASSP 81; Mar. 30, 31, Apr 1, 1981; Atlanta, Ga., pp 240-242.
2. G. F. Boudreaux and T. W. Parks, "Thinning Digital Filters: A Piecewise-Exponential Approximation Approach", IEEE Trans., Vol. ASSSP-31, No. 1, Feb. 1983, pp. 105-113.
3. M. V. Thomas, Y. Neuvo, and S. K. Mitra, "Two-Dimensional Interpolated FIR Filters", Proc. ISCAS 83, May 2-4, 1983; Newport Beach, CA, pp. 904-906.
4. Y. Neuvo, D. Cheng-yu, and S. K. Mitra; "Interpolated Finite Impulse Response Filters", IEEE Trans. Acoustics, Speech, and Signal Processing, Vol. ASSP-32, No. 3; June 1984, pp. 563-570.
Different studies have suggested that, in most instances, one may omit from a finite-impulse-response filter having multiple coefficient multipliers one out of two coefficients, three out of four coefficients, or seven out of eight coefficients. Interpolation of the filter output restores the signal. Nevertheless, the same number of delay elements on the input delay line for the filter is necessary so the tap coefficients that remain in the filter properly operate on the input signal and combine in the summer at the proper time.
Each tap weight or coefficient multiplier 32 of the thinned filter is an analog value, as in the conventional analog FIR filter. Such analog multipliers are acceptable for low-frequency signals, such as voice, but are inappropriate for high-frequency signals in the gigahertz range because of the difficulty in accurate multiplication, as discussed above. Thus, these decimated or thinned filters have been used in telephone and other low-frequency systems; but, the problems of multiplying the input signals by the analog coefficients accurately enough and fast enough for high-frequency sampled-data signals such as are used in high-frequency communication systems have prevented their use there. In addition, the problems associated with rapid changes of the filter coefficients in agile filters remain in the thinned or decimated filter.
The present invention is a signal processor useful with high-frequency signals, such as high-frequency sampled-data signals, for filtering those signals accurately, and providing the capability of changing the filter coefficients or characteristics quickly and with accuracy.
The invention is a hybrid charge-transfer-device filter including a thinned finite-impulse-response filter in which each filter coefficient of a thinned filter is expressed or approximated as a multibit digital word. Each bit of this multibit digital word forms a tap weight on a tap from the input delay line of the filter. The output of each tap weight is scaled by a value corresponding to the significance of that tap weight bit in the multibit word, and is delayed so the outputs of the taps forming a single word are combined together to form the filter output.
The signal processor of the invention includes a delay line having a plurality of delay elements. The signal processor further includes bit-tap elements, each having an input coupled to one of the delay elements to multiply the tap-element input by a single bit of a multibit word comprising a filter coefficient of a thinned FIR filter. The signal processor further includes a summer for combining the outputs of the bit-tap elements, and a plurality of delay elements coupled to the outputs of at least some of the bit-tap elements to delay those outputs so the outputs of the bit-tap elements forming a single filter coefficient are combined together in the summer into a single processor output.
FIG. 1 is a schematic diagram of a conventional analog finite-impulse-response filter of the prior art.
FIG. 2 is a schematic diagram of a finite-impulse-response filter using binary-analog correlators.
FIG. 3 is a schematic diagram of a thinned finite-impulse-response filter.
FIG. 4 is a schematic diagram of one embodiment of the hybrid filter of the invention.
FIG. 5 is a model of a CCD cell for use in modeling a filter.
FIG. 6 is a graph illustrating filter leakage for various values of fixed-pattern noise (FPN), and charge-transfer inefficiency (CTI), superimposed upon a parametric plot of FIR filter parameters that relate the number of taps, the transition width, the passband ripple, and the peak stopband gain for the agile bandpass filter.
FIG. 7 is a graph illustrating the effects of frequency-dependent-weighting errors due to charge-transfer inefficiency mismatch.
FIG. 8 is a graph illustrating noise gain as a function of charge-transfer-inefficiency mismatch.
In an analog thinned or decimated FIR filter such as that shown in FIG. 3, only a few of the tap locations on the input delay line 34 are used. In the signal processor of the present invention, a thinned finite-impulse-response discrete-time analog filter has each of its filter coefficients expressed as a multibit digital word. Each bit is implemented as a bit-tap weight coupled to an input delay line at the locations liberated by the thinning of the analog filter to produce a hybrid charge-transfer device filter structure. This hybrid filter takes analog input data, and applies digital-filter coefficients to that data. Such a hybrid filter is both more accurate and easier to manufacture than a conventional variable analog-tap-weighted FIR CCD filter.
The hybrid filter of the invention preferably includes an input delay line having on it a plurality of delay elements. In the embodiment of the present invention shown in FIG. 4, the delay line 40 has seven delay elements 42. The representation shown is for illustration only. As will be apparent to those skilled in the art, a typical high-performance, high-frequency signal filter will have significantly more delay elements and potential tap locations, perhaps up to several hundred. Nevertheless, the principles of the present invention may be applied to a filter of any size.
In the filter of the invention a plurality of binary tap weights or binary-coefficient multipliers 44 are coupled to taps on the input delay line 40. Together these binary-coefficient multipliers form the coefficient multipliers of a thinned or decimated finite-impulse-response filter in which only certain of the coefficient multipliers of a conventional FIR filter are used. The thinned-filter coefficients are determined using known thinned or decimated analog finite-impulse-response filter techniques.
Each analog filter coefficient of the thinned filter is expressed in binary form as a multibit word. In the embodiment shown in FIG. 4, each filter-coefficient multiplier is expressed as four-bit word. Nevertheless, the number of bits used to express each coefficient of the thinned filter depends upon the desired degree of specificity. Although a four-bit-per-coefficient implementation is shown for illustration, the coefficients may typically be represented by binary words having up to fourteen, or perhaps more, bits.
A filter tap weight coefficient, ak, may be expressed as an N-bit number in binary form: ##EQU1## where the αkn are the bits (0, 1) that form ak, αko is the sign bit, and |ak |<1.
Statistically, each αkn is 0 or 1 with equal likelihood. Therefore, during any given clock interval half of the information bits are unused, to the detriment of the signal-to-noise ratio of the device. The Buttner-Schussler form can be used to obtain bits having values of ±1.
The negative of ak is formed by complementing all bits and adding a "1" in the least significant bit (LSB) position, i.e., ##EQU2## Now, ak can be expressed: ##EQU3## where
βko =-(αko -αko) (5)
βkn =(αkn -αkn), n≠0, N-1 (6)
βk, N-1=(αk, N-1 -αk, N-1)-1 (7)
The coefficient ak is thus expressed by coefficients having values ±1, except for the least significant bit, which has values 0 or -2.
Each bit of the multibit filter coefficient is used as a bit-tap weight 44 on a tap from the input delay line 40. The tap weight multiplies the tap input by the value of the tap weight. In this way, each tap and its corresponding bit-tap multiplier correspond to one bit of a multibit thinned-filter coefficient. In the filter embodiment shown in FIG. 4, each tap is taken from the output of one of the input line delay elements 42. An additional control tap is taken from the input of the first delay element. Thus, the filter has N taps from a delay line having N-1 delay elements 42.
The bit multipliers or weights of the binary-coefficients 44 are preferably plus or minus one (±1). Such a binary multiplier is easy to implement and is accurate in high-frequency applications because of the ease with which an input signal can be multiplied by a unitary coefficient. In addition, the use of plus or minus one, rather than one or zero, provides a better signal-to-noise ratio in the output from the multiplier by supplying more definite outputs.
In the example shown, the first tap weight 44a, 44e of each group forming a four-bit-coefficient multiplier is representative of the least-significant bit of the four-bit word. The first tap having the first binary tap weight 44a is, in the example shown, coupled to the input to the first delay element 42a of the input delay line. The next bit-tap weight 44b is coupled to the output of the first delay element 42a or the input to the second delay element 42b. That tap weight is preferably representative of the next-least-significant bit of the thinned-filter coefficient. That construction is continued so that the Nth tap, with its Nth-bit tap weight 44d forming the most-significant bit of the N bit word forming the filter coefficient is taken from the output of the (N-1)th delay element 42c of the delay line.
For a filter having more than one multiplier coefficient expressed in binary form, the delay element, tap, and bit-multiplier-weight structure is repeated for each filter-multiplier coefficient. In the filter structure shown in FIG. 4, two multiplier coefficients are each represented by four-bit words. Thus, each coefficient is represented by four individual single-bit-weight multipliers. The bit-tap weights for each filter coefficient are preferably coupled to the delay line in the same order for each coefficient. It will be noted that the number of taps divided by the thinning factor is an integer.
At a particular time, for a signal input to the signal processor of FIG. 4, the signal at the input to the first delay element 42a and tapped by the first tap element of the first coefficient word is designated xn+3. The input to the second delay element 42b is xn+2 ; the input to the third delay element 42c is xn+1 ; and the input to the fourth delay element 42d is xn. The input to the fifth delay element 42e, which is tapped by the first tap element of the second coefficient word, is designated xn-1 ; the input to the sixth delay element 42f is xn-2 ; the input to the seventh delay element 42g is xn-3 ; and the output from the seventh delay element 42g is xn-4.
In the illustrated embodiment the first and fifth bit-factor elements 44a, 44e represent the least-significant bits of the two thinned-filter coefficients a0 and a1. According to the calculations above, these bit multipliers may each have a value of 0 or -2. Thus, the first coefficient is (-1/2) β03 and the fifth coefficient is (-1/2)β13. The outputs 46a, 46e of those multipliers are therefore (-1/2)β03 xn+3 and (-1/2)β13 xn-1, respectively.
The second and sixth bit-factor multipliers 44b, 44f represent the next-least-significant bits of the coefficients a0 and a1. Those bit multipliers have values β02 and β12, respectively, each of which may equal ±1. The respective outputs 46b, 46f of those multipliers are β02 xn+2 and β12 xn-2.
Similarly, the third and seventh bit-factor multipliers 44c, 44g represent the second-most-significant bits of their respective four-bit coefficients. Those bit multipliers have values β01 and β11, respectively, each of which may equal ±1. The respective outputs 46c, 46g of those multipliers are β03 xn+1 and β11 xn-3.
Finally, the fourth and eighth bit-factor multipliers 44d, 44h represent the most-significant bits of the two four-bit coefficients. Those multipliers have values β00 and β10, respectively, which are ±1. The respective outputs 46d, 46h of those multipliers are β00 xn and β10 xn-4.
Coefficient-multiplier bits having the same significance in their respective coefficients are subsequently factored and combined to form the output, processed signal. The outputs of bit-tap weights having the same significance in each of their respective coefficients are combined in an adder that combines the outputs of the bit-tap weights having common significance in the multibit coefficient words. The outputs of all the bit-taps forming a single coefficient word are combined together to form a multibit output signal word.
The outputs from the bit multipliers representing the least-significant bits of the four-bit coefficients ak are combined in a first adder 48 to produce an analog equivalent of a least-significant-bit output 49 of: ##EQU4##
The outputs from the bit multipliers representing the next-least-significant bits are combined in a second adder 50 to produce an analog equivalent of a next-least-significant bit output 51 of: ##EQU5##
The outputs from the bit multipliers representing the next-most-significant bits are combined in a third adder 52 to produce a next-most-significant bit output 53 of: ##EQU6##
The outputs from the bit multipliers representing the most-significant bits are combined in a fourth adder 54 to produce an analog equivalent of a most-significant bit output 55 of: ##EQU7##
Because the inputs to the bit-tap weights 44 for the different bits of a single filter coefficient are delayed different amounts in the input delay line 40 prior to being tapped off and multiplied by the bit-tap weights, the outputs of the taps must be properly delayed before combination to ensure that all the output bits associated with a single multibit coefficient-multiplier word are combined together at the correct time to produce an output that is the same as would be obtained if the input had been directed through a single coefficient multiplier having a value equal to the value represented by the multibit word of the bit-tap weights.
The next-least-significant bit output 51 from the second and sixth bit-factor multipliers is one unit of time delayed with respect to the least-significant bit output 49 from the first and fifth bit-factor multipliers because of the first delay element 42a on the input delay line. Therefore, the least-significant-bit output 49 should be delayed one unit of time in a first tap-output delay element 60 prior to combination in a first summer 62 with the next-least-significant-bit output 51.
The bit-tap-weight outputs should also be appropriately weighted prior to combination in the summer to maintain the proper relative significance of the different tap outputs in the combined result. But, in the illustrated embodiment the least-significant-bit multipliers 44a, 44e have as factor of 1/2 built in, so no additional weighting is necessary prior to combining the least-significant and next-least-significant-bit outputs 49, 51. Because of the sign attached to the bit factors β03 and β13, the least-significant-bit output 49 is input as a negative into the first output summer 62.
The output 64 from the first output summer 62 is a signal representative of the least-significant and next-least-significant bits of the tap-weight output.
The combined output 53 from the third and seventh bit multipliers 44c, 44g are one unit of time delayed with respect to the output from the second and sixth multipliers 44b, 44f, and hence are also one unit of time delayed from the output 64 from the first output summer representing the least-significant and next-least significant bits of the tap-weight output. Therefore, the combined output 64 from the first output summer is delayed in a second output delay element 66 before being combined with the outputs 53 from the third and seventh bit multipliers. Also, since the output from the third and seventh bit multipliers is to have greater significance in the final output than the output from the first output summer, the output 64 from the first output summer is scaled by 1/2 before being combined with the output of the third and seventh bit multipliers in a second output summer 68.
Similarly, the output 70 of the second output summer 68 is scaled by 1/2 and delayed one unit of time in a third output-delay element 72 before being combined in a third output summer 74 with the combined output 55 of the fourth and eighth bit-tap weight multipliers. The resulting signal on the summer output 76 is represented by: ##EQU8## Referring back to Equation 4, we see this is: ##EQU9## Thus, the signal processor constructed as described and shown has a gain factor of 2 built into it.
After combination, an interpolator 78 acts on the output 76 to fill in the missing parts of the signal lost due to dropping some of the coefficients and to restore the output signal. Such interpolators are known in the art in connection with conventional analog thinned or decimated FIR filters. Such interpolation restores the missing points on the impulse curve with adequate accuracy, as discussed in the articles previously referred to. In some applications, it has been found that interpolation as simple as a straight line yields acceptable results.
Because of the simplicity of changing the bit-tap weights for the binary-coefficient multipliers in the filter of the invention, it is easy to change the coefficient multipliers in the filter just described. In addition, because of the binary nature of each bit-tap weight, the coefficient multipliers may be changed accurately and quickly, without the need for a great deal of adjustment to obtain accurate analog-filter coefficients. In addition, the binary-coefficients of each bit-tap weight allow accurate, high-speed multiplication of high-frequency input signals.
Many of the advantages of the present invention can be appreciated by comparing the performance of the filter using a plurality of parallel binary-analog correlators as shown in FIG. 2 with the embodiment of the invention shown in FIG. 4.
FIG. 5 illustrates a model of the kth CCD cell in a string of CCD cells as part of the input-delay line. The input from the previous delay-element cell is xk-1, and the output to the succeeding cell is xk. Because of feedback and charge-transfer inefficiency (CTI, η), the delay element does not operate as an ideal element, but operates essentially in accordance with the model illustrated in FIG. 5. The sensed output from the delay element yk is input to the tap weight coupled to that delay element.
It has been found that the effective tap weight coupled to each sensed output is composed of a weighted sum of the actual tap weight and the preceding tap weight. Consequently, charge-transfer inefficiency results at each cell.
In the binary-analog-correlator filter, the transfer function of each correlator is a function of a number of factors. An actual transfer function differs from the theoretical ideal transfer function, leaving an error transfer function as the difference between the actual and ideal transfer functions.
Each tap element has its own charge-transfer inefficiency η. The CTI, η, of the various taps are not likely to be identical, so an inconsistency Δη of charge-transfer inefficiencies exists. That inconsistency contributes to filter "leaks". Noise also arises as fixed-pattern noise (FPN) due to inaccuracies in reading out the charge from each well of the transfer device.
FIG. 6 illustrates the filter leaks for various fixed pattern noise (Δf) and charge-transfer inefficiency (CTI) inconsistencies Δη superimposed upon a parametric plot of finite-impulse-response filter parameters that relate the number of taps, the transition width, the passband ripple, and peak stopband gain for the agile bandpass filter.
Weighting errors cause another independent filter leak, the effect of which is shown in FIG. 7. The curves illustrated in the left half of the graph of FIG. 7 can be converted to represent the curves for other values of σ. To convert the 1 σ curves to 2 σ or 4 σ curves, the 1 σ curves are shifted one bit or two bits, respectively. To remain accurate, the filter should maintain the frequency-dependent weighting error less than 1/2 the value of the least-significant bit.
Noise gain then becomes a function related to the log of the square of the charge-transfer-inefficiency inconsistency, which is the difference among the charge-transfer inefficiencies of the various delay elements of the correlator. The total noise gain for the correlator-filter output is plotted in FIG. 8. The number M represents the number of tap weights.
Using the embodiment of the invention shown in FIG. 4, the number of delay elements and taps is substantially reduced from that of the binary-analog-correlator filter shown in FIG. 2. As a result, the filter will tolerate a larger inconsistency or mismatch among the charge-transfer inefficiencies of the various delay lines or delay elements and still maintain acceptable overall performance. This can be seen from FIG. 8, in which it is seen that the fewer the number of taps, the lower the noise gain for a given charge-transfer-inefficiency inconsistency.
This analysis and conclusion are based upon White, "Apparent Performance Limit of a Binary-Analog Correlator (BAC) CCD Filter," Proc. 1983 IEEE International Symposium on Circuits and Systems, Newport Beach, CA; May 24, 1983; pp. 1354-1357.
|Cited Patent||Filing date||Publication date||Applicant||Title|
|US3676654 *||May 21, 1970||Jul 11, 1972||Collins Radio Co||Digitalized filter|
|US4149259 *||May 12, 1978||Apr 10, 1979||U.S. Philips Corporation||Transversal filter for convoluted image reconstruction|
|US4495591 *||Mar 26, 1984||Jan 22, 1985||The Regeants Of The University Of California||Pipelined digital filters|
|US4507725 *||Jul 1, 1982||Mar 26, 1985||Rca Corporation||Digital filter overflow sensor|
|US4524423 *||Nov 6, 1981||Jun 18, 1985||Rca Corporation||Digital signal separation filters|
|US4542475 *||Nov 23, 1982||Sep 17, 1985||Rca Corporation||Sampled data filter system as for a digital TV|
|US4611305 *||Aug 17, 1983||Sep 9, 1986||Sony Corporation||Digital signal processing circuit|
|US4641259 *||Jan 23, 1984||Feb 3, 1987||The Board Of Trustees Of The Leland Stanford Junior University||Adaptive signal processing array with suppession of coherent and non-coherent interferring signals|
|US4644488 *||Oct 12, 1983||Feb 17, 1987||California Institute Of Technology||Pipeline active filter utilizing a booth type multiplier|
|US4691292 *||Apr 13, 1983||Sep 1, 1987||Rca Corporation||System for digital multiband filtering|
|US4691293 *||Dec 28, 1984||Sep 1, 1987||Ford Aerospace & Communications Corporation||High frequency, wide range FIR filter|
|US4701874 *||Dec 12, 1983||Oct 20, 1987||Nec Corporation||Digital signal processing apparatus|
|Citing Patent||Filing date||Publication date||Applicant||Title|
|US4918637 *||Sep 26, 1989||Apr 17, 1990||Hewlett-Packard Company||Multichannel decimation/interpolation filter|
|US5117291 *||Nov 30, 1990||May 26, 1992||At&T Bell Laboratories||Technique for adjusting signal dispersion cancellation apparatus in communications systems|
|US5130942 *||Oct 17, 1991||Jul 14, 1992||Canon Kabushiki Kaisha||Digital filter with front stage division|
|US5233549 *||Apr 21, 1992||Aug 3, 1993||Loral Aerospace Corp.||Reduced quantization error FIR filter|
|US5367476 *||Mar 16, 1993||Nov 22, 1994||Dsc Communications Corporation||Finite impulse response digital filter|
|US5404322 *||Mar 27, 1992||Apr 4, 1995||Seiko Corp.||Digital filter and method of design|
|US5590154 *||Jun 26, 1995||Dec 31, 1996||Motorola, Inc.||Equalizer circuit and a method for equalizing a continuous signal|
|US5898878 *||Mar 25, 1998||Apr 27, 1999||Sony Corporation||Data processing system having capability to interpolate processing coefficients|
|US5999954 *||May 9, 1997||Dec 7, 1999||Massachusetts Institute Of Technology||Low-power digital filtering utilizing adaptive approximate filtering|
|US6209013||Dec 13, 1999||Mar 27, 2001||Altocom, Inc.||Efficient implementation of a filter|
|US6618739||Feb 22, 2001||Sep 9, 2003||Altocom, Inc.||Digital filter implementation suitable for execution, together with application code, on a same processor|
|US7212052 *||Aug 25, 2004||May 1, 2007||Samsung Electronics Co., Ltd.||Jitter suppressing delay locked loop circuits and related methods|
|US7263542 *||Apr 29, 2002||Aug 28, 2007||Koninklijke Philips Electronics N.V.||Inverse filtering method, synthesis filtering method, inverse filter device, synthesis filter device and devices comprising such filter devices|
|US7398288||Aug 29, 2003||Jul 8, 2008||Broadcom Corporation||Efficient implementation of a filter|
|US20040136268 *||Apr 29, 2002||Jul 15, 2004||Den Brinker Albertus Cornelis||Inverse filtering method, synthesis filtering method, inverse filter device, synthesis filter device and devices comprising such filter devices|
|US20050088211 *||Aug 25, 2004||Apr 28, 2005||Kim Kyu-Hyoun||Jitter suppressing delay locked loop circuits and related methods|
|WO1991001526A1 *||Jul 13, 1990||Feb 7, 1991||At&E Corporation||Digital filter and method of design|
|U.S. Classification||708/319, 708/306|
|Oct 10, 1985||AS||Assignment|
Owner name: ROCKWELL INTERNATIONAL CORPORATION,
Free format text: ASSIGNMENT OF ASSIGNORS INTEREST.;ASSIGNOR:WHITE, STANLEY;REEL/FRAME:004464/0699
Effective date: 19850821
|Jul 25, 1989||CC||Certificate of correction|
|May 4, 1992||FPAY||Fee payment|
Year of fee payment: 4
|Aug 28, 1996||FPAY||Fee payment|
Year of fee payment: 8
|Oct 8, 1996||REMI||Maintenance fee reminder mailed|
|Jan 14, 1999||AS||Assignment|
Owner name: CREDIT SUISSE FIRST BOSTON, NEW YORK
Free format text: SECURITY INTEREST;ASSIGNORS:CONEXANT SYSTEMS, INC.;BROOKTREE CORPORATION;BROOKTREE WORLDWIDE SALES CORPORATION;AND OTHERS;REEL/FRAME:009719/0537
Effective date: 19981221
|Dec 15, 1999||AS||Assignment|
Owner name: CONEXANT SYSTEMS, INC., CALIFORNIA
Free format text: ASSIGNMENT OF ASSIGNORS INTEREST;ASSIGNOR:ROCKWELL SCIENCE CENTER, LLC;REEL/FRAME:010415/0761
Effective date: 19981210
|Aug 25, 2000||FPAY||Fee payment|
Year of fee payment: 12
|Oct 3, 2000||AS||Assignment|
Owner name: BOEING COMPANY, THE, CALIFORNIA
Free format text: MERGER;ASSIGNORS:ROCKWELL INTERNATIONAL CORPORATION;BOEING NORTH AMERICAN, INC.;REEL/FRAME:011164/0426;SIGNING DATES FROM 19961206 TO 19991230
|Nov 5, 2001||AS||Assignment|
Owner name: CONEXANT SYSTEMS, INC., CALIFORNIA
Free format text: RELEASE OF SECURITY INTEREST;ASSIGNOR:CREDIT SUISSE FIRST BOSTON;REEL/FRAME:012252/0413
Effective date: 20011018
Owner name: BROOKTREE CORPORATION, CALIFORNIA
Free format text: RELEASE OF SECURITY INTEREST;ASSIGNOR:CREDIT SUISSE FIRST BOSTON;REEL/FRAME:012252/0413
Effective date: 20011018
Owner name: BROOKTREE WORLDWIDE SALES CORPORATION, CALIFORNIA
Free format text: RELEASE OF SECURITY INTEREST;ASSIGNOR:CREDIT SUISSE FIRST BOSTON;REEL/FRAME:012252/0413
Effective date: 20011018
Owner name: CONEXANT SYSTEMS WORLDWIDE, INC., CALIFORNIA
Effective date: 20011018